Recovering structure from many low-information 2-D images of randomly-oriented samples
New sources and detectors are allowing scientists to look at matter with finer spatial and temporal resolutions. These experiments can produce data that are a series of severely Poisson limited snap-shots of randomly oriented samples. An extreme case of this is destructive imaging of single particles with an X-ray free-electron laser-many frames are needed for a reconstruction, but there is no a priori information associated with the frames about particle orientation. We use Cornell's Pixel Array Detectors (PADs) to examine the practical limits of an expectation maximization (EM) algorithm designed to deal with extremely low-fluence data, having just a few photons per frame. We demonstrate image reconstruction of a high-contrast sample using hundreds of thousands of randomly oriented frames with an average X-ray photon occupancy as low as 2.5 photons per frame. Practical aspects of reducing low-fluence data, such as thresholding and noise limits, will be discussed for high- and low-contrast samples; and data collected in the presence of significant background signal.